Brokers and vendors are seemingly offering clients more algorithmic products than Kellogg's has cornflakes. An exaggeration? Of course, but you get the idea. The buyside, however, might just feel that way, as some providers expand their algorithmic offerings, while others are still getting into the game. With a multitude of offerings on its doorstep, the buyside may have entered a frazzled state of mind perhaps best described as "algo fatigue."
Still, the buyside trader does see value in algorithms. They've become a necessity for a number of reasons. Those range from protecting anonymity, to minimizing market impact, to lowering commissions. Besides broker and vendor offerings, the buyside can develop their own algorithms, though most don't.
But how is a buyside firm supposed to know which broker's algorithms work best? Are they all the same? Is one better than another?
Transaction cost analysis (TCA) might be one way to find out. TCA analysis firms such as Quantitative Services Group (QSG), ITG's Plexus Group, Abel/Noser Corp. and Elkins/ McSherry are beginning to test the quality of various brokers' algorithms, so that buyside users can compare and contrast strategies.
EdgeTrade is one of the few broker-dealers to offer up its algos for comparison with those of its peers. The small agency brokerage contracted with Elkins/McSherry to do TCA on its VWAP algorithm.
VWAP Analysis
"When it comes to best execution and analyzing trade data, it is not easy to get your arms around it. We wanted an independent party to verify how we are doing," Kyle Zasky, president of EdgeTrade, says.
Elkins/McSherry's analysis showed that EdgeTrade's VWAP outperformed 93 percent of a universe of large institutions including asset managers and funds during 2005. Zasky believes TCA will grow in importance in the near future: "There is me too' and catchup going on now because of the pressure to justify where you are sending your order flow. You don't want any fingers pointed at you by customers. We are pretty bullish on the TCA idea."
TCA for algorithms is not a one-size-fits-all solution, its critics maintain, and is still in its infancy. TCA itself has been around for years, and is an integral part of the trading process. And it is the industry's acceptance of this post-trade critique that is helping lead to the concept of measuring the effectiveness of algorithms.
Costs associated with a transaction go well beyond the explicit-such as commissions, markups and fees-to implicit costs, which are also critical. Market impact, in particular, is becoming the one cost the buyside is paying closer attention to when analyzing algorithmic strategies.
Susquehanna Financial Group (SFG) was another broker-dealer to look to a measurement firm for feedback on its algos.
Susquehanna asked Quantitative Services Group (QSG) to test its algorithmic strategies against its peers, using SFG's data from one quarter. QSG found that SFG beat competitors by a total of 18.5 basis points in terms of total execution cost, largely because of its ability to minimize market impact.
David Margulies, head of program and algorithmic sales at Susquehanna Financial Group, says SFG commissioned the report for internal purposes: "We wanted to see how we did versus our peers, so we could tweak the algorithms or hit different liquidity pools."
Trumpets Results
Because the report looked so favorably upon Susquehanna's algorithms, the broker-dealer decided to trumpet its success with a marketing campaign. "It is very hard for the buyside to tell what is working and what is not working. I think it is a fair analysis because QSG is the last independent execution analysis firm; there is no skew to the results," Margulies says.
Margulies is right. Four of the five trade-cost measurement firms are owned by companies that offer brokerage. Elkins McSherry is the property of State Street; Abel/Noser provides trading services; Plexus Group is owned by ITG; and BECS is owned by Citigroup.
Slow on the Uptake
John Wightkin, managing partner at QSG, says he is not sure algorithmic TCA is a trend as such, just a void that needs to be filled in the push toward getting the buyside to trade more frequently with algorithms.
Randy Grossman, research manager in the capital markets group at Financial Insights, agrees. He says TCA is one of the top drivers in getting the buyside to take up algorithmic trading. It was number six in his report on the "Top 10 Capital Markets Initiatives for 2006."
"The race for algorithmic trading supremacy continues on the transaction cost analysis front," Grossman writes. "Firms must be able to offer sophisticated analytics that assist fund managers in choosing algorithms and analyzing the impact of those trades to compete successfully."
Wightkin sees an opportunity in the marketplace for QSG to give the buyside more information on broker algorithms. He believes the inability to measure broker algos has hampered their acceptance somewhat on the buyside. Wightkin stressed, however, that measuring broker algorithms is in its infancy.
Brian Fagen, managing director and U.S. head of electronic sales at Lehman Brothers, differs in that he believes that algorithmic trading by the buyside has actually become an accepted part of the business in a relatively short period of time. "At the beginning, it was surprising how quickly some took it up."
Fagen does agree that measuring algorithms is important "so people can see what the results are, but comparing apples to apples is relatively difficult."
Adam Sussman, a consultant at the TABB Group, says he doesn't see a lack of TCA holding the buyside back from using algorithms: "They want to know how well the algorithms are performing, but there is not an overwhelming cry from the buyside for TCA that compares broker algorithms against each other."
Ian Domowitz is the CEO of ITG Solutions Network, the subsidiary that houses ITG's TCA analytical products and Plexus Group.
"My sense is that the measurement business is growing; there is increasing concern as the order flow increases," says Domowitz. He says as algorithmic trading becomes more ingrained on the buyside, the performance of the algorithm is more important than the features and functions.
ITG will do a comparative report on algorithmic performance on a strategy-by-strategy basis or a vendor-by-vendor basis, depending upon what a buyside client requests, says Domowitz. However, he does not see a great deal of interest in performance analysis from the sellside at this point.
Apples and Oranges
Comparing algos to algos, so to speak, is complicated, where every broker's algorithm is designed to be better, faster and more cost-efficient.
Domowitz notes that the issue is not just with algorithmic trades: "It existed long before algorithms. Everyone has different strategies in use, even if they are not automated."
Wightkin says if one has the right metrics to use in measuring, "apples to apples" is achievable. "We have metrics that give us the ability to measure across algorithms," he says. "Using VWAP and TWAP you can standardize, measure and categorize; you can begin to get apples to apples."
He notes there is a continuum between not measuring TCA at all and getting it exactly right. The answer is somewhere in between. Simply measuring against a benchmark, however, "doesn't give you enough insight into what is going on under the surface. It won't tell us how it did or the costs associated with that execution." QSG addresses this with a "forensic" way of analyzing trade costs, Wightkin says.
The Debate
Fagen disagrees: "Measurement should be based on how well traders do versus a benchmark, not the algorithm. A good trader is still critical. The best algorithm in the world can never make up for a bad trading decision."
Ted Oberhaus, director of equity trading at Lord Abbett & Co., which manages $76 billion in equities, says he doesn't see how ranking broker algorithms benefits him or his firm.
Buyside traders are constantly adjusting an algorithm or moving to another algorithm as they alter their strategies on an order. Consequently, a trading strategy isn't a static thing that begins at the open and ends at the close, he says. That makes measuring or ranking a broker's algorithm difficult. He says that TCA in the scenario he described is actually analyzing the trader's decision-making, not the algorithm.
"The algo is what the trader makes of it," says Oberhaus. "You can't measure the tool itself. If you're overlaying manual intervention over an algorithm, you couldn't utilize a TCA vendor to measure a broker's algorithm."
Sussman concludes: "It is the parameters the buyside sets that affect the algorithms and how they behave. In measuring performance a lot of miscalculations can occur. You might get more information on the aptitude of the clients and how they behave when they trade algorithmically than you get on how the actual algo performs."
Wightkin stands by his product.
"If the buyside is going to dip its toes into algorithmic trading, it has to be able to compare and rank them. I've got to believe that there are discussions at a senior level where they are evaluating different brokers, making sure they are not getting picked off," he says.
QSG is currently working on a broader product, says Wightkin, a kind of regularly updated league table of algorithm providers that could be available as early as the autumn. But Wightkin says he doesn't think brokers like the idea of TCA testing their algorithms.
Post Trade
Sellside firms may prefer to let the buyside do its own analysis. Tom Bok, senior vice president in the equity quantitative analytics group at Lehman Brothers, adds that TCA measurements should be done firm by firm. "The buyside needs to make post-trade analysis an integral part of its overall process," Bok says. "This needs to be done individually and thoughtfully by each buyside firm."
Even if a trader tweaks the parameters of an algorithm, says Domowitz, the endgame is still to beat or achieve the benchmark. "There are slightly different ways to get to a benchmark, but you can still get there," Domowitz says. "This is not an excuse not to measure algorithms."
Wall Street thrives on rankings. It has a league table for everything, from measuring M&A activity to the underwriting of various investment instruments. Could it also have a league table to measure the effectiveness of brokerage algorithms?
Seal of Approval
It wouldn't hurt marketing, says TABB's Sussman. A high ranking in such a league table, he says, would in effect provide an endorsement of that algorithmic strategy, and the use of that broker's product would likely spike.
Unfortunately, it's not that simple. At least not yet, according to Fagen and Bok of Lehman Brothers. They are not convinced that a league table comparing algorithms is possible at this time, nor that would it be accurate.
Fagen says, "The fact that the data is coming from an order management system means that it may not be tracking the start or the end time properly. That makes it difficult to know what the measurement means. There is a lot of work to be done on the data management side."
Data can be an issue, agrees Wightkin: "It is a big universe, and we are scraping together the tags that tell us what algorithm they used, the start time, the end time. It is just starting to become more available."
ITG's Domowitz says that data tagging has gotten considerably better, but still needs improvement: "The algorithm is tagged, the vendor is tagged. The problem may lie with the strategy, which is not often tagged."
Data Issues
One source familiar with TCA agrees with Domowitz, saying that clients' OMSs currently don't capture the data necessary to rank the algorithmic products of brokers and vendors. "Clients are not looking for this type of granularity, even if they could capture the data," the source says.
Bok works in Lehman's quant group, and he says a league table would be good in one sense: It would raise the awareness and acceptance of post-trade analytics. However, a league table would be difficult to produce because there is too much happening around the post-trade data to measure it objectively.
"You need to measure the trader in partnership with the algo," Bok explains. "A league table approach is too blunt of an instrument for this process. It has to be done by the buyside firm, within the context of the firm's investment process."
Sussman adds, "A VWAP bake-off is not what the industry needs. It needs strategies to meet or beat the benchmarks the algorithms were designed to implement."
_________________________
